The limited approvals can ease immediate pressure on local hyperscalers but leave longer-term constraints in place, with implications for global enterprise AI infrastructure.
China has begun cautiously reopening access to Nvidia’s advanced AI processors, granting approvals to a small group of its largest technology companies.
The companies include ByteDance, Alibaba, and Tencent, which are expected to collectively purchase more than 400,000 of Nvidia’s H200 accelerators, according to Reuters. Other technology firms may line up for approvals in subsequent rounds.
[ Related: More Nvidia news and insights ]
The approvals signal a shift by Beijing as it balances immediate AI compute demand against its longer-term mandate to build a domestic semiconductor ecosystem.
“This unlocks a predictable supply of high‑performance accelerators, enabling large Chinese enterprises to plan denser GPU clusters, higher‑bandwidth fabrics, and faster model‑training roadmaps,” said Charlie Dai, VP and principal analyst at Forrester. “Hyperscalers can further speed up scaling AI infrastructure pipelines and optimizing for multi‑petaflop training capacity rather than relying solely on uneven domestic alternatives.”
Impact on AI workloads
The approvals may force Chinese hyperscalers to rethink how AI workloads are distributed.
“We expect leading Chinese enterprises to adopt a dual-track deployment strategy: prioritizing H200s for core large-scale model training workloads, while allocating domestic chips primarily for inference and smaller-scale training tasks,” said Galen Zeng, senior research manager for semiconductor research at IDC Asia/Pacific.
China is also accelerating efforts to strengthen domestic training chip design and manufacturing capabilities, with the strategic objective of reducing long-term dependence on foreign suppliers, Zeng added.
Things could get more complex if authorities mandated imported chips to be deployed alongside domestically produced accelerators. Reuters has reported that this may be a possibility.
“A mandated bundling requirement would create a heterogeneous computing environment that significantly increases system complexity,” Zeng said. “Performance inconsistencies and communication protocol disparities across different chip architectures would elevate O&M [operations and maintenance] overhead and introduce additional network latency.”
However, the approvals are unlikely to close the gap with US hyperscalers, Zeng said, noting that the H200 remains one generation behind Nvidia’s Blackwell architecture and that approved volumes fall well short of China’s overall demand.
Implications for global enterprises
For global enterprise IT and network leaders, the move adds another variable to long-term AI infrastructure planning.
Expanded sales of Nvidia’s H200 chips could help the company increase production scale, potentially creating room to ease pricing for Western enterprises deploying H200-based AI infrastructure, said Neil Shah, VP for research at Counterpoint Research.
“If increased volumes allow Nvidia to improve manufacturing efficiency and supply predictability, Western enterprises could see more flexibility in pricing and availability as they continue rolling out H200-based AI systems,” Shah said. “Any cost efficiencies could also help offset rising memory prices for some Nvidia customers, if the company chooses to pass those savings on.”
Any enterprises experimenting with Chinese open-source AI models could also see indirect benefits if access to Nvidia’s H200 chips allows those models to advance more quickly in capability and scale, Shah added. Zeng said Beijing may expand H200 procurement quotas over time, while Nvidia and AMD are expected to introduce China-specific AI chip variants later this year to comply with export controls.
More Nvidia news:
- Nvidia is still working with suppliers on RAM chips for Rubin
- RISC-V chip designer SiFive integrates Nvidia NVLink Fusion to power AI data centers
- Nvidia H200 chips in China: US says yes, China says no
- Lenovo-Nvidia partnership targets faster AI infrastructure rollouts
- Top 10 Nvidia stories of 2025 – From the data center to the AI factory
- HPE loads up AI networking portfolio, strengthens Nvidia, AMD partnerships
- Nvidia’s $2B Synopsys stake tests independence of open AI interconnect standard
- Nvidia bets on open infrastructure for the agentic AI era with Nemotron 3
- Nvidia moves deeper into AI infrastructure with SchedMD acquisition
- Nvidia chips sold out? Cut back on AI plans, or look elsewhere
- Nvidia’s first exascale system is the 4th fastest supercomputer in the world
- Nvidia highlights considerable science-based supercomputing efforts
- Nvidia’s first exascale system is the 4th fastest supercomputer in the world
- Nvidia touts next-gen quantum computing interconnects
